Recognition, Based On Orthogonal Fourier - Mellin Moments Of The Digital Symbol

Nowadays in the field of Geographic Information System(GIS), it is a problem eagerly needing to solve to use automatic vectorization of scanned map to replace time-killed traditional manual digitalization. However, the key technology of automatic digitalization is mainly automatic recognition of the important data in the map, including the recognition of digital annotations in the scanned map.In this paper, the Orthogonal Fourier-mellin Moments (OFMMs) is extracted as the feature. Compared with the seven invariant moments from Hu's method, OFMMs is more accurate, more robust to the noise, more efficient to reconstruct signal, and has no information redundancy. After we extract the OFMMs of the digital annotations, these features will be input into the high order neural network(HONN) as train data, which has no hidden layers at all and has little consumption of computation. This presented method is not sensitive to the change produced by rotation and rescale, at the same time goes fast.According to the test result from 20 GIS maps, the accurate rate of the recognition is 90% or so. Experimental result shows that it is efficient to use neural network based on HONN to recognize number, which can be put into practice.